Get help when PlayWeld is not working
Start with what you were doing and find the right guide for setup, projects, engines, AI, or backups.
On this page
When PlayWeld does not work as expected, start with the last step that went well and the first one that did not. The desktop app, local service, game engine, AI provider, and website account each play a different part.
The desktop application will not start
Check the exact package's operating-system requirements and notes. For a source checkout, use its pinned dependencies and native host toolchain. Keep Windows and Linux dependency directories separate. See building from source for the starting commands.
Project Home cannot reach the service
Confirm that both application and CLI use the intended profile. From a built checkout, run node packages/platform-service/lib/cli.js status with that profile selected. Inspect startup and endpoint diagnostics before deleting lock or authentication files. Closing the desktop may leave the service running.
Project or engine actions are missing
Open the intended registered workspace and inspect the screen's project selector. Confirm native files under game/, the correct installed executable, and the named capability layer. A live editor connection must identify the correct game before mutation.
See engine setup and project organization.
A model route or agent task fails
For an eligibility error, check enabled models, capabilities, relevant pools, context limits, and budgets. For a provider failure, inspect the endpoint response and account configuration. Questions, approvals, and blocked dependencies are different waiting states; examine the existing task before creating another request.
Knowledge, packages, or recovery need attention
For stale search, inspect pending work and reconcile before a source-filtered lexical check. For packages, check type, compatibility, activation, and runtime prerequisites. For a restore, use a verified archive and a fresh empty destination; redacted settings cannot replace credential recovery.
Collect useful evidence
Record version, host, project, operation and task/run/call IDs, reproduction steps, and exact error. Preserve the failing artifact and remove credentials and private content before sharing. Audit and diagnostics explains how to connect the visible problem to persisted records.